Saturday, February 1

DeepSeek hit with large-scale cyberattack, says it’s limiting registrations

Due to widespread malicious attacks on its services, DeepSeek said on Monday that it would temporarily restrict user registrations. However, current users will still be able to log in normally.

As a rapidly expanding competitor to OpenAI’s ChatGPT, Google’s Gemini, and other top AI tools, the Chinese AI firm has created a lot of excitement in recent weeks.

Earlier Monday, DeepSeek overtook rival OpenAI as the most downloaded free software in the United States on the Apple software Store, unseating ChatGPT in favor of DeepSeek’s AI Assistant. It contributed to a major sell-off in international tech equities.

Tech analysts, investors, and developers have heard about the company, which was created in 2023 and debuted its R1 model last week. They believe that the buzz and the resulting fear of falling behind in the constantly shifting AI hype cycle may be justified. Particularly in the age of the generative AI arms race, when startups and IT behemoths are vying for market share in a field expected to generate over $1 trillion in revenue in the next ten years.

With a focus on large language models and achieving artificial general intelligence (AGI), a branch of AI that equals or surpasses human intellect on a wide range of tasks, which OpenAI and its competitors claim they are rapidly pursuing, DeepSeek reportedly emerged from a Chinese hedge fund’s AI research unit in April 2023.

Particularly after the startup’s publication of R1, its reasoning model that competes with OpenAI’s o1, last week, the buzz surrounding DeepSeek started to grow. Users have praised its performance and reasoning abilities, and it has soared to the top of app stores and industry leaderboards. It is open-source, so any AI developer can utilize it.

See also  Black Friday Apple deals: Get discounted AirPods, MacBooks and more

Despite the U.S. halting chip exports to China three times in three years, the startup’s models were noteworthy. The precise cost of DeepSeek’s R1 and the number of GPUs used are subject to variation in estimates. A newer version, according to Jefferies analysts, only cost US$5.6 million to train (assuming US$2/H800 hour leasing rate). That is less than ten percent of Metas Llama’s price.

Nevertheless, reports concur that the model was created at a fraction of the price of competing models from OpenAI, Anthropic, Google, and others, regardless of the precise figures.

More from CNBC:

  • Trump tariffs could raise prices on technology like laptops, smartphones and AI

  • Why this China-made BYD Shark pickup is drawing attention in the global truck market

  • Today is the first day you can file your 2024 taxes 4 key updates to know before filing

The AI industry is therefore beset with concerns, such as whether a bubble is poised to pop and whether the industry’s growing number of outrageous financing rounds and billion-dollar valuations is required.

Leave a Reply

Your email address will not be published. Required fields are marked *